0
$\begingroup$

I need to calculate the variance of $$ E = \sqrt {(X - a)^2 + (Y -b)^2} - Z ~, $$ where $X$ and $Y$ are Normally distributed variables with zero mean and $Z$ is Log-Normally distributed. All $X, Y$, and $Z$ are uncorrelated with known mean and variances, while $a$ and $b$ are constants.

I am very much new to this area but trying calculate variance of $E$. My questions are as follows:

  1. Is the following correct? $$ \mathrm{Var}(E) = \mathrm{Var}\left( \sqrt {(X - a)^2 + (Y -b)^2}\right ) + \mathrm{Var}(Z)~.$$ This identity is true for two independent Normal distributed variables but I doubt that this will be true for this case as well.
  2. Can somebody point out some resource for calculating the variance of square root terms?
$\endgroup$
  • $\begingroup$ To make a square root of a big expression, use \sqrt{expression} with curly braces. To make left and right parentheses than look nice, use \left( and \right). $\endgroup$ – 6005 Oct 24 '16 at 3:39
0
$\begingroup$

The answer to your first question is yes: Because $X$, $Y$, $Z$ are independent, then $W = ((X-a)^2 + (Y-b)^2)^{1/2}$ is independent of $Z$, hence $$\operatorname{Var}[E] = \operatorname{Var}[W] + \operatorname{Var}[Z].$$

The answer to your second question is harder. We can observe, however, $$\operatorname{Var}[W] = \operatorname{E}[W^2] - \operatorname{E}[W]^2,$$ and the expectation of $W^2$ is actually easy, because $(X-a)^2 + (Y-b)^2 \ge 0$ for all $X, Y, a, b$. Linearity of expectation gives us $$\operatorname{E}[W^2] = \operatorname{E}[(X-a)^2] + \operatorname{E}[(Y-b)^2],$$ and since $X$ is normal with mean $0$ (you did not specify the variances of $X$ and $Y$), we have $$\operatorname{E}[(X-a)^2] = \operatorname{E}[X^2] - 2a \operatorname{E}[X] + a^2 = \operatorname{Var}[X] + \operatorname{E}[X]^2 - 0 + a^2 = \sigma_X^2 + a^2,$$ where $\sigma_X^2$ is the variance of $X$. It follows that $$\operatorname{E}[W^2] = \sigma_X^2 + \sigma_Y^2 + a^2 + b^2.$$ The expectation of $W$ itself is more problematic; at this time, I don't have an elegant solution; I can only suggest directly evaluating the integral $$\operatorname{E}[W] = \int_{x=-\infty}^\infty \int_{y=-\infty}^\infty ((x-a)^2 + (y-b)^2)^{1/2} \frac{e^{-x^2/(2\sigma_X^2)} e^{-y^2/(2\sigma_Y^2)}}{2\pi \sigma_X \sigma_Y} \, dy \, dx.$$ But I suspect this integral does not have an elementary closed form. I doubt it is integrable even when $\sigma_X = \sigma_Y = 1$.

$\endgroup$
  • $\begingroup$ Thanks for such a clear answer. i will evaluate integration part. $\endgroup$ – vicky Oct 24 '16 at 5:23
  • $\begingroup$ yes! you are right this E[W] integral does not turn out be closed form . Is there any way to approximate this function which can give me a closed form of integral . $\endgroup$ – vicky Oct 24 '16 at 7:40
  • $\begingroup$ thanks for comments. please give me any direction on the issue mentioned in above comment $\endgroup$ – vicky Oct 24 '16 at 7:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.